Near-infrared Spectroscopy Combined With Chemometrics Applied Research In The Rapid Detection Of Milk

Milk powder is nutrient-rich, containing nutrients which are necessary for growth and development of human and metabolism, moreover, it is easily digested and absorbed. The detection of the major nutrients in milk powder still depends on chemical methods currently. Most determinations of quality control are destructive, expensive, time and labor consuming, and of-line by nature, which can't satisfy the requirement of modern quality control.Nowadays, facing ever-increasing quality problem of milk powder, more and more people have realized that it is important to supervise timely the quality of milk powders.So, it is urgent to develop a simple and fast determining method.Compared the routine chemical methods, near-infrared diffuse reflectance spectroscopy has advantages such as high speed, simultaneous non-destructive measurement of a number of constituents. In the paper, in combination with the chemometrical methods, the Fourier transform near-infrared (FT-NIR) spectroscopy analysis technology was used for rapid detection of milk powder, in order to provide a reference for fast, rea-ltime and non-destructive determination of milk powder quality. The main contents and conclusions are as follows:1. In the introduction part of this thesis, the function and quality control status of milk powder, the common chemical detection methods of the key indicators composition of protein and fat and its application in milk powder; the principle, the development overview, the application features, the use and the research progress in dairy products of near-infrared spectroscopy are summarized.2. The chemometric methods in analysis of near infrared spectroscopy, including four parts:pretreatment methods of spectroscopy data, quantitative calibration, qualitative methods of pattern recognition and model transfer. In addition, division of calibration samples and model evaluation are also contained.3.100 milk powder samples of many different brands, varieties and production batches, and 150 milk powder amples including 50 infant formulas,50 youth milk and 50 middle-aged milk sare collected for this study. The contents of proetein and fat, which the two main nutrition ingredients of milk powder were selected as detecting indexes. The chemical values of 100 milk powder samples, which were determined by the conventional chemical analyze methods of Kjeldahl and Rose-Gettlied separately, were taken as the accomplished reference data for quantity prediction matrix. The near-infrared spectrum of all milk powder samples were scanned by Antaris II Fourier transforms near-infrared spectrometer. The author groped the best parameter enactments of spectropmeter by means of choosing and revising scanning parameter of spectropmeter, samples status, and process methods of samples. The near-infrared spectrum of milk powder samples were scanned accordingly. The samples'spectrum scanning parameters were established as follows. The scope of scanning wave number: 10000cm-1-4000cm-1; the resolution:4cm-1; the background and samples scanning times:64; the detector:InGaAs. The rotational sample receptacle was revolved to reduce the sample non-uniform influence.4. Near-infrared spectra (NIRS) technique combined with support vector machines (SVM) was used to identify three varieties of different age rank milk powder.Firstly,120 training set was selected among total 150 samples using Kennard-Stone method. The multiple scatter correction (MSC) was used for eliminating scattered light. Radical basis function as kernel function, two model parameters including kernel functionγand penalty factor C were Optimized by two-step grid searching and five fold cross validation, the best value ofγand C was 0.03125 and 2048, respectively. The best parameters were used to build qualitative correction model, the accuracy of the training set and test set both could reach 100% with the built model. The result showed that the discrimination of different age rank milk powder by SVM is feasible.5. Near infrared (NIR) spectroscopy combined with chemometric methods was used for establishing a new method to determine the content of fat and protein in milk powder. The training and testing sets were partitioned by Kernard-Stone algorithm. Wavelet transformation (WT) was used for de-noising and compressing to signal. The models were built by radical basis function neural networks (RBFNN) after spectral signal reconstruction. The parameter Spread value of RBF network, the wavelet functions and decomposition levels were discussed in detail. The wavelet function, compression level and spread value was 3.5, db2 and fourth for fat, respectively, Correlation coefficient obtained of testing set (Rp) is 0.990, Root mean square error prediction(RMSEP) is 0.007. The above-mentioned three parameters were db8, fourth and 6 for protein, respectively, Rp is 0.994, RMSEP is 0.004. The results showed that the RBFNN combined WT are used for building NIR models and predicting, the model is much more representative, robust, and the prediction accuracy is improved. This method is suitable for determining the fat and protein of milk powder rapidly and nondestructively.